A BYY Split-and-Merge EM Algorithm for Gaussian Mixture Learning

Gaussian mixture is a powerful statistic tool and has been widely used in the fields of information processing and data analysis. However, its model selection, i.e., the selection of number of Gaussians in the mixture, is still a difficult problem. Fortunately; the new established Bayesian Ying-Yang (BYY) harmony function becomes an efficient criterion for model selection oil the Gaussian mixture modeling. In this paper, we propose a BYY split-and-merge EM algorithm for Gaussian mixture to maximize the BYY harmony function by splitting or merging the unsuited Gaussians in the estimated mixture obtained from the EM algorithm in each time dynamically. It is demonstrated well by the experiments that this BYY split-and-merge EM algorithm can make both model selection and parameter estimation efficiently for the Gaussian mixture modeling.